X2 4450e tested on Abit AN-M2: very low power

All about them.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
Palindroman
Posts: 247
Joined: Fri Oct 20, 2006 1:00 am

X2 4450e tested on Abit AN-M2: very low power

Post by Palindroman » Sun Apr 13, 2008 12:49 am

I've had my eye on the Abit AN-M2 (nVidia 7025/630) for quite a while. I knew that it was one of the most efficient boards around, right behind the Asrock Alive NF7G-HDready. Unfortunately I don't have the money or the time to test every board around. Besides, I was very happy with the Gigabyte boards I use to build low power systems for my clients, albeit they consume slightly more power.

But I was lucky this time, doubly so. Due to the meagre availability of Gigabyte boards I had a chance to order the Abit AN-M2 for a server and test it. On top of that my client wanted a dualcore CPU, preferably a BE-2400. I convinced him to go for the brand new X2 4450-e.

Here are the power consumption figures:

Image

edit: In front of those figures it should say
RMClock idle
Cool 'n Quiet
Idle
RMClock load
Load (Orthos)


As you can see, 20.3 Watts on idle is pretty low. That's undervolted, 1.0 GHz @ 0.850 V. I managed to undervolt the CPU all the way down to 0.762 V (1 hour Orthos) and get 19.7 W. However, when after that I tried to find minimal voltage for the 2.3 GHz stock speed the system would crash continuously. Only at 1.0 GHz @ 0.850 V and 2.3 GHz@ 1.100 V the system remained more or less stable (taking the voltages a notch up would probably be best). I've had this problem with RMClock and BE-2xx0 chips as well. Does anyone know what's up with these Brisbanes?

To put things in perspective, here's the Abit/4450e compared to a Gigabyte/LE-1100:

Image


edit: In front of those figures it should say
RMClock idle
Cool 'n Quiet
Idle
RMClock load
Load (Orthos)

The difference is mainly due to the motherboards, quite spectacular at idle too IMO, considering the difference in CPU performance. I must say that I really liked this Abit board. It was better than the Gigabyte in most respects (notably BIOS and heatsink temps), except for one: dynamic fan control. With the EasyTune utility I can turn off the CPU fan completely and when it kicks in at idle, if at all after an hour or so, it does so at 600-1000 rpm which makes it practically inaudible. With Abit's FanEQ I couldn't get the fan below 1500 rpm. It's the only thing keeping this board from perfection IMO.

To top it off I have some performance results (PCMark05):

Image

edit: In front of those figures it should say
System
CPU
Memory
Graphics
HDD

I hope someone can use these test results to his or her advantage. If anyone knows how to undervolt those Brisbanes all the way down with RMClock, or how to get Abit's FanEQ make that CPU fan go below 1000 rpm, I'm all ears.
Last edited by Palindroman on Sun Apr 13, 2008 7:01 am, edited 1 time in total.

Charuto
Posts: 43
Joined: Sat Mar 17, 2007 10:46 pm
Location: USA

Post by Charuto » Sun Apr 13, 2008 1:42 am

Thanks for this. I like seeing mention of mainboard power consumption.

plympton
Posts: 229
Joined: Sun Mar 14, 2004 11:40 am

Brisbanes

Post by plympton » Sun Apr 13, 2008 6:42 am

I bought a Black Edition 5000+ x2 and had the same crashing with it and RMClock. I ended up selling the 5000+ since I didn't have time to futz with it. It was really unfortunate, since I wanted the ability to go up to 3 GHz when I needed it (encoding), but clock at 2 GHz normally, and 1 GHz at idle without the fear of instability caused by raising the FSB.

If I stuck it at low volts at 3 GHz, it would be rock solid. If I did the dynamic scaling, crash-o-matic. This was on Vista Home Premium, but it's an otherwise rock-solid system running Media Center & Slimserver, AnyDVD (mobile) encoding with a EE 3800+ x2.

If I have to get rid of the MB in order to fix this, I figure I might as well get an Intel setup for the next machine since they are pretty competitive power-wise these days, but probably will wait a year or so at this point.

-Dan

alphabetbackward
Posts: 107
Joined: Sat Apr 15, 2006 9:59 pm
Location: California

Re: Brisbanes

Post by alphabetbackward » Sun Apr 13, 2008 7:07 am

plympton wrote:I bought a Black Edition 5000+ x2 and had the same crashing with it and RMClock. I ended up selling the 5000+ since I didn't have time to futz with it. It was really unfortunate, since I wanted the ability to go up to 3 GHz when I needed it (encoding), but clock at 2 GHz normally, and 1 GHz at idle without the fear of instability caused by raising the FSB.

If I stuck it at low volts at 3 GHz, it would be rock solid. If I did the dynamic scaling, crash-o-matic. This was on Vista Home Premium, but it's an otherwise rock-solid system running Media Center & Slimserver, AnyDVD (mobile) encoding with a EE 3800+ x2.

If I have to get rid of the MB in order to fix this, I figure I might as well get an Intel setup for the next machine since they are pretty competitive power-wise these days, but probably will wait a year or so at this point.

-Dan
I just recently got a Black Edition with a Gigabyte 690G board. I found I could get it as low as 5x with a low undervolt but anytime I wanted it to go back to a higher state, it crashed. At the moment, I have it set for Maximal Performance at 15x at 1.325 (stock voltage) and Power Savings at 8x at 1.075. Over time, I'm going to experiment with some more undervolting or different multipliers. I decided not to do the dynamic scaling as it seems to be a bit unstable but that might be a voltage issue.

So 3.0 GHz when I need the performance, 1.6 GHz when not.

plympton
Posts: 229
Joined: Sun Mar 14, 2004 11:40 am

Re: Brisbanes

Post by plympton » Sun Apr 13, 2008 8:19 am

alphabetbackward wrote:So 3.0 GHz when I need the performance, 1.6 GHz when not.
That's a bummer. I don't think it's a voltage thing - it's either an RMClock bug, or (worse) a BIOS/CPU issue. It's not a supported way to use the CPU, I'm sure, and RMClock is, at best, a great hack.

Since I'm using an HTPC, manually switching between modes isn't really an option (it's an old blurry TV, and I can barely navigate in Windows as it is...

-Dan

Palindroman
Posts: 247
Joined: Fri Oct 20, 2006 1:00 am

Post by Palindroman » Sun Apr 13, 2008 9:16 am

I didn't have time to fiddle around with CrystalCPUID but I definitely will with the next Brisbane I get my hands on. I'll go to the Rightmark forums and ask again, because it really is a shame the undervolting with RMClock doesn't work as well as it used to with previous Brisbanes and still does with Sparta Semprons. They're very focussed on Santa Rosa lately, not so much on AMD I feel. Undervolting is most popular with notebook owners after all.

RedAE102
Friend of SPCR
Posts: 118
Joined: Wed Jul 04, 2007 3:29 pm
Location: Lost and Found Bin, Cypress, CA, USA
Contact:

Post by RedAE102 » Sun Apr 13, 2008 10:31 am

I had to give up running RMClock on either of my Brisbanes. I have a G1 3600+ that was a less than willing undervolter that eventually became highly unstable, so it's no longer in use. I also bought a 2.8 GHz 5400+ G2 Brisbane that I am now running in my Biostar TF7050-M2. For automatic clock speeds, I use CrystalCPUID, and over the past few weeks, I have yet to experience a crash from it. I did have to make sure I set the timeouts for 200ms and the wait time to 50ms, and only allow single step transitions: Min-->Mid-->Max and Max-->Mid-->Min. Oh, and RMClock doesn't work on my friend's Tyler Turion TL-56 either. It's a 1.8 GHz 65nm dual core, same as Brisbane. My undervolting results on this 5400+ chip are as follows:
0.775V @ 1.2 GHz (0.768V CPU-Z) (Min usable VID)
0.950V @ 2.0 GHz (0.936-0.944V CPU-Z)
1.075V @ 2.4 GHz (1.056-1.072V CPU-Z)
1.225V @ 2.8 GHz (1.200-1.216V CPU-Z) (1.325V is stock)

My G1 3600+ could only manage:
0.800V @ 800 MHz
1.075V @ 1.9 GHz
1.225V @ 2.4 GHz (overclocked: 253x9.5)

A few notes on this... Selecting single voltages in RMClock, the range of 0.550V-1.400V was available (0.550-0.775 in 0.0125V steps, and 0.775-1.400V in 0.025V steps), but selecting the 0.7625V, 0.750V, or anything lower than that resulted in it jumping back up to 1.550V and then stepping down from that. It's a shame, since this runs at 0.775V even at 1.2 GHz, which indicates it could go lower at 1GHz or 800MHz given the chance. Also, CrystalCPUID only shows 0.800V-1.550V as my available voltage range, with the usual question marks outside of the 1.100V-1.350V range. I got around this by figuring out the numbers that CrystalCPUID assigns the VIDs... it starts at 0=1.550V and goes inversely up from there... 9=1.325V 10=1.300V 11=1.275V up to 31=0.775V. Then it was only a matter of replacing the value CqMinVID=31 in the CrystalCPUID.ini settings file using Notepad. If anyone wants more clarification on this stuff, let me know, and I'll make a new thread on it with more detailed information and screenshots.
Last edited by RedAE102 on Mon Apr 14, 2008 6:40 pm, edited 1 time in total.

walkingjohn
Posts: 29
Joined: Sun Mar 18, 2007 9:51 am
Location: Ontario, Canada

Post by walkingjohn » Sun Apr 13, 2008 4:11 pm

Those having rmclock/brisbane troubles, have you fiddled with the "management" settings in rmclock? Or the transition intervals?

I set mine to "perform single-step transitions only" (with a few intermediate p-states in the on-demand profile) and that seemed to help--but it has been a while since I've fiddled with it so I don't remember the details. My 3600+ is stable now between [email protected] and [email protected]. I believe I too backed off lower voltages a bit to avoid crashes.

Bembotto
Posts: 8
Joined: Fri Apr 25, 2008 2:33 am
Location: Italy

Post by Bembotto » Fri Apr 25, 2008 5:25 am

Doesn't asrock 7050 chipset mobo require the 12V pin? I know pico psu hasn't this 4pin plug...Is it right?

loimlo
Posts: 762
Joined: Thu May 03, 2007 3:58 am
Location: Formosa

Post by loimlo » Fri Apr 25, 2008 9:48 am

I think every boards in the market require a ATX12V plug. You can buy an ATX12V adapter for Pico if you want, and many people have done it before.

Bembotto
Posts: 8
Joined: Fri Apr 25, 2008 2:33 am
Location: Italy

Post by Bembotto » Fri Apr 25, 2008 9:54 am

loimlo wrote:I think every boards in the market require a ATX12V plug. You can buy an ATX12V adapter for Pico if you want, and many people have done it before.
What is? Where i can buy it? Have you a link?
Thnak you

ryboto
Friend of SPCR
Posts: 1439
Joined: Tue Dec 14, 2004 4:06 pm
Location: New Hampshire, US
Contact:

Post by ryboto » Fri Apr 25, 2008 10:15 am

Bembotto wrote:
What is? Where i can buy it? Have you a link?
Thnak you
Go to Short-circuit.com, they sell the Pico and any adapters you might need for a pretty good price.

PenDragon
Posts: 41
Joined: Mon Feb 16, 2004 5:31 pm

Post by PenDragon » Thu May 01, 2008 11:39 am

Just wanted to throw my hat in the ring here with the 5000+ Black Edition and RMClock...

I tried RMClock out without reading any instructions, and was getting instant-power-off crashes within 5 minutes of hitting Apply to the Power On Demand profile. After reading through this thread and a couple of other pages, I checked out the P-States and VID's for the 5000+ BE to see what it is spec'ed to run at. Based on what I found, I made some changes to the settings, and so far, it's been completely stable *fingers crossed*.

here's the settings that I used:
I disabled direct multi-state transitions
Based on the processors specs, I chose the following P-states/voltages
1.0ghz @ 1.1v
2.0ghz @ 1.15v
2.4ghz @ 1.25v
3.0ghz @ 1.35v

That's all the settings that I used, and so far it's been stable.

The reasoning behind the settings is that the crashes would occur during P-State transitions, so I'm guessing it was the voltage deltas that were crashing it. So, I'm only allowing one P-state transition at a time, and specifically, every P-state transition has a delta of .1v maximum. Since I disabled direct multi-state transitions, thus slowing down the throttling speed, I decided to use as few steps as possible while keeping the voltage delta at .1v or below.

Now I realize that these settings don't give me any more power savings than CnQ would, BUT it's still a huge benefit to me, because the moment you change the multiplier in BIOS on an AMD system from Default to anything else at all (only available with an unlocked CPU such as mine), CnQ is immediately disabled. So, I had to choose between power savings (system idles at 89w) without the overclock to 3.0Ghz, or an overclock to 3.0Ghz without the power savings (125w at idle) on a Media Center that is on 24/7. Now, I get the best of both worlds!

I will be trying to lower the voltages a bit if I get continued stability with current settings, maybe starting at 1.0v instead of 1.1 and going up from there.

davidh44
Posts: 54
Joined: Wed Apr 16, 2008 8:51 pm
Location: USA

Post by davidh44 » Thu May 01, 2008 3:43 pm

RedAE102 wrote:For automatic clock speeds, I use CrystalCPUID, and over the past few weeks, I have yet to experience a crash from it. I did have to make sure I set the timeouts for 200ms and the wait time to 50ms, and only allow single step transitions: Min-->Mid-->Max and Max-->Mid-->Min.
Did CrystalCPUID randomly crash/reboot/BSOD like RMClock does when you didn't have those exact settings?
If anyone wants more clarification on this stuff, let me know, and I'll make a new thread on it with more detailed information and screenshots.
Yes, please!

RedAE102
Friend of SPCR
Posts: 118
Joined: Wed Jul 04, 2007 3:29 pm
Location: Lost and Found Bin, Cypress, CA, USA
Contact:

Post by RedAE102 » Fri May 02, 2008 12:11 am

For the followers of this thread, I offer this:

viewtopic.php?p=410542

I've run a cumulative 11 days on the current settings with no crashes. I can't in good conscience leave a computer running for more than 8 days without a reboot. Hence, it's actually 8 days, plus 3 days since my own self imposed reboot. Unfortunately, the settings I had posted earlier in this thread did result in a greyish pattern screen of death that all previous attempts have. So far, so good with the settings in my new thread. Knock on wood. Seriously, I did just get up and knock on my oak dresser. Yes, I know, I'm ridiculous, but we all have our sticking points.

astrayan
Posts: 45
Joined: Fri Jun 17, 2005 5:54 am
Location: astraya

Post by astrayan » Thu May 08, 2008 6:14 am

PenDragon wrote:here's the settings that I used:
I disabled direct multi-state transitions
Based on the processors specs, I chose the following P-states/voltages
1.0ghz @ 1.1v
2.0ghz @ 1.15v
2.4ghz @ 1.25v
3.0ghz @ 1.35v

That's all the settings that I used, and so far it's been stable.
No point?. I was recently playing with a 5000+ @ 2.6Ghz, and it was stable near 1.1V.

This thread is so disappointing. Now I'm back to square one. I thought the instability was from the motherboard M2A-VM (690G). It got substantially more unstable when I installed the ATI chipset drivers. I tried single state transitions on the integer multipliers, but after the driver install, RMClock wouldn't even get through boot, and it became a major game to sequence things so that it could drop through each "delta V" to reach 0.8V. Once you got there, you could lock it in to a "Power Saving" profile, but what a pain.

It took me about half an hour to figure out RMClock's weird new interface which had changed so much since 2005.

The good news is Palindroman's Abit motherboard 20.3W discovery, which means it doesn't matter so much. I find the undervolted top power for the chips a bit strange at 70W, since mine was about +30W under load.

thejamppa
Posts: 3142
Joined: Mon Feb 26, 2007 9:20 am
Location: Missing in Finnish wilderness, howling to moon with wolf brethren and walking with brother bears
Contact:

Post by thejamppa » Fri May 09, 2008 2:01 pm

Excellent news. I just c hecked local retail price for 4450e in here: 61€'s. Oh, yeah this baby is going to replicate my old 65W 4000+. I just updated my Bios for that ^^

Anyway, this 4450e seems really sweet low power CPU. Since I am always upgrading to lower my power consumption, this hits the sweet spot!

jojo4u
Posts: 806
Joined: Sat Dec 14, 2002 7:00 am
Location: Germany

Post by jojo4u » Sat May 10, 2008 10:46 am

Nice findings, Palindroman. The Asrock NF7G-HDReady consumes 17 to 18 W as you mentioned in another post in the gallery? What's the most efficient Intel-based setup at the moment? I still remember the Abit iL-90MV here ;)

Post Reply